Changelog

All notable changes to MEME are documented here. The format follows Keep a Changelog; versions follow Semantic Versioning, and git tags carry the v prefix Go modules require (v0.1.0-alpha).

Unreleased

Added

• docs/roadmap.md: the coverage roadmap. It records the 2026-10-25 measurement of what each framework accepts natively against what the emitters write for the whole scenario and example corpus, lists the emitter code paths no payload exercises, names two defects the capability reference hides (PyPSA storage units lose capacity.max and operation.min_pu/max_pu; AdOpT batteries ignore storage.charge_eff/discharge_eff), and plans the additive work for 0.2.x and the breaking bundle for 0.3.0, each with its deprecation. Technology catalogs become selectable: a library pin on a technology or arc names an upstream entry (AdOpT database, PyPSA standard types) or asks for a MEME-synthesized definition, and experiment.catalog sets the default.

0.2.0-alpha - 2026-10-25

Added

• Power flow as a PyPSA run mode (ADR-0008): experiment.mode: power_flow solves the AC network for a fixed dispatch with lpf() and pf() instead of optimizing. The canonical model describes the network itself — node.electrical.v_nom, transmission.electrical (kind line or transformer, r, x, s_nom, num_parallel, tap_ratio) and technology.dispatch (overridable per node) — and experiment.power_flow names the slack, the method (newton_raphson or linear), distribute_slack and the tolerance. The run writes output/power_flow/: convergence.csv, per-bus voltage magnitude, angle and power, per-line and per-transformer flows and loading, generator dispatch and a summary.json; a snapshot that does not converge fails the run. The API consumer chains a plan and a power flow as two calls; MEME holds no state between them.
• Gates around the new inputs: the mode routes through the power_flow capability (PyPSA only); power-flow inputs in an optimization mode (a dispatch, the electrical blocks) are kept and named in one warning while the arcs stay transport links (see Changed); in mode power_flow expandable capacities, trade, constraints, emission limits and scenarios are rejected, and distribute_slack requires a dispatch on the slack technology (PyPSA shares the balance in proportion to each generator's set-point, so a slack without one would receive none of it).

• Scenario pypsa_power_flow, the example payloads examples/pypsa_power_flow*.json (the reference network, the smallest valid power flow, a meshed ring with per-node dispatch, and the options: linear method, distributed slack, storage sign, a second carrier) and E2E checks for convergence, the linear method and a diverging load. The examples walkthrough documents the options, the result files and the rejected requests.

• Reactive power, voltage control, shunts and controlled transfers for the power flow (ADR-0009): technology.electrical with reactive_dispatch or power_factor (+ leading), control: pv and voltage_setpoint; node.electrical.shunt; transmission.dispatch as a controlled HVDC transfer on an arc without impedance; b, g, phase_shift and r_per_length/x_per_length on transmission.electrical. The result folder gains the reactive tables, the load and link tables and the shunt tables. Gates: one way to give Q, voltage set-points for pv generators and the slack only and agreeing per bus, an electricity arc needs an impedance or a transfer, every AC island needs a generator (an island without the named slack is reported as balanced by PyPSA), and the linear method says what it ignores. Scenario and example payload pypsa_power_flow_reactive, E2E-pinned on the held voltages.

• The electrical network inside the optimization (ADR-0010): experiment.network.model: linear_power_flow makes PyPSA solve under Kirchhoff's voltage law with Lines and Transformers (expansion from the arc's capacity, electrical.max_loading as s_max_pu), and network.security.outages runs the security-constrained solve; other targets reject the model like a mode. Transmission gains lifetime, interest_rate and cost_basis, and PyPSA prices the arc's investment and fixed O&M on Links and Lines. Every optimization run writes output/results/ (dispatch, flows, storage states, marginal prices, capacities.csv, statistics.csv, summary.json).

• PyPSA optimization completeness (ADR-0011): an energy-sized storage (storage.energy_capacity) is a Store with charger and discharger Links, which honors independent rates, depth of discharge and the energy-capacity cost (_storage.json carries the equal-power and rate constraints); storage.soc_profile pins a StorageUnit's state of charge; capacity.per_unit builds integer modules; decommission.mode: continuous retires existing capacity at its fixed O&M and builds new capacity above it; allow_unmet_demand adds priced slack generators; copperplate joins every node through a free hub; operation.ramp_start_up, ramp_shut_down, standby_cost, initial_uptime, initial_downtime and costs.<class>.variable_om_quadratic map onto PyPSA's unit-commitment and cost attributes; model.periods with years are investment periods with build_year, lifetime and new builds per period. Scenarios and goldens for every claim, E2E on the real framework, and the example examples/pypsa_planning.json.

Changed

• PyPSA no longer rejects independent storage rates, depth of discharge, continuous retirement, unmet demand or copperplate; the energy-capacity cost is priced instead of dropped. Calliope and AdOpT-NET0 keep every mapping they had; the new fields they have no slot for (soc_profile, the unit-commitment extras, the quadratic cost, investment periods) are ignored there with a warning. The validation matrix records every verdict that moved.
• time.subset and costs.<class>.purchase are rejected on PyPSA and AdOpT-NET0 instead of being dropped without a word: neither target has a slot (narrow time.start/end; fold the purchase cost into investment_per_capacity). Calliope 0.7 warns that area.per_capacity is not emitted (0.6 emits resource_area_per_energy_cap). The validation matrix records the moved verdicts.
• Power-flow inputs — dispatch, the electrical blocks on technologies, arcs and nodes — are ignored with one warning wherever no power flow runs: on Calliope and AdOpT-NET0 and in PyPSA's optimization modes alike. A dispatch in an optimization mode was rejected before; now one payload serves the planning call and the power-flow call (ADR-0009).
• The PyPSA driver parses its run configuration as JSON (json.loads) instead of pasting the JSON in as a Python literal, which broke on the first boolean setting.

Fixed

• PyPSA state_of_charge_initial is absolute (MWh), but the canonical fraction storage.initial_soc was written into it unchanged, so a half-full 24 MWh unit started at 0.5 MWh. The fraction is now applied to the installed energy (p_nom × max_hours); an expandable storage is told it starts from its existing capacity.

0.1.0-alpha - 2026-10-24

The first pre-release. It carries the service as scaffolded in July 2026 — the canonical model and its validation, the emission toolkit, the target registry with capability-driven validation, the PyPSA, Calliope and AdOpT-NET0 targets, the job execution engine with bounded concurrency, the HTTP API with multi-target resolution and API-key authentication, the embedded scenario corpus with golden snapshots and the validation matrix, the containerized solver environment, the real-solver E2E suite, the published JSON Schemas and the example payloads — together with everything that landed on it since, listed below. Before 1.0 the API and the canonical model may still change between releases.

Added

• meme -version and the version field of GET /healthz report the build version, which make build stamps from the git tag.
• Calliope 0.6.8 as the calliope-0.6 target: its own emitter for the 0.6 model definition (model/run/techs/locations/links, file= timeseries, group_constraints, conversion_plus carrier tiers), its own capability profile (k-means typical days and an emissions-minimizing objective gained; indexed parameters, piecewise performance and custom math absent) and native key native["calliope-0.6"] (#4).
• Framework versions as separate targets: a default per framework, version variants selected by name, version-explicit aliases (calliope-0.7), and GET /capabilities reporting default_targets, aliases and versions.
• Environment image with one virtual environment per Calliope release, provisioned by uv (Calliope 0.6.8 runs on a managed Python 3.8).
• Scenarios and cross-framework E2E checks for multi-node technologies (calliope_multi_node, calliope_multi_node_supply) and operate-mode limits (calliope_operate_limits); the same payload is pinned to the same objective on PyPSA, Calliope 0.7 and Calliope 0.6.
• A CI workflow (format, vet, race tests, build, schema check, vulnerability scan) and a strict documentation build deployed to GitHub Pages.
• Validation of timeseries.source (inline with values, file with a path) and rejection of a node listed twice on one technology.

Changed

• Calliope 0.7 target migrated from 0.7.0.dev7 to the stable 0.7.0 release: data tables are spelled table / add_dims.inputs, a supply technology's fuel cost is cost_source_use, ratio-pinning math no longer uses the removed defined() helper, and node coordinates are emitted on all nodes or none (#3).
• Calliope data tables are named after technology and parameter (<tech>__<param>.csv) instead of the series id two technologies could share.
• The documentation moved from one long README into a MkDocs site; the capability reference lives at docs/reference/capability-reference.md.

Removed

• The organisation-template workflows sync-legal.yml and sync-branch-rules.yml with their scripts: run from this repository they acted on every repository of two GitHub organisations and needed the template's tokens, so every run here failed. The template-only pages on legal sync and branch rules left the documentation site; the branch-naming and commit-convention pages stay, reworded for MEME, and the lint scripts point at them.

Fixed

• Calliope: a technology placed at several nodes wrote its demand profile, availability or source series for the first node only; every placed node now gets its own data-table column, and per-node profile overrides land in their node's column (#7).
• Calliope: in operate mode a trade import ran unbounded and a transmission line ignored its capacity; both are installed at their fixed size, as are source_cap and area_use, with a warning where capacity.max stands in for the installed size, and a model with no fixed capacity is rejected before the solve instead of aborting after it (#8). The same gap was closed on the 0.6 target, where the import ran above its limit.
• Calliope: a transmission line's capacity.existing is honored in plan mode (it used to be ignored, leaving a fixed line unbounded).
• Calliope: an already-annualized cost basis pins cost_depreciation_rate: 1 instead of being divided by the lifetime again.
• PyPSA: a payload without inline series is solved over the whole model.time range instead of a single default snapshot.
• A file-backed time series, which no target can materialize, is rejected at validation instead of failing at emit time.
• Container: the API service restarts unless stopped; development tooling corrected and tsam pinned below 4 for AdOpT-NET0.
• Earlier fixes: CORS validation for credentialed wildcards and reading the CORS setting from the environment; solver process-group cleanup, output capping and bundle symlinks; capability gates on node overrides; numeric edge cases, resolution parsing and YAML floats in the emitters; a node with no technologies emitted as techs: {}.